Handwritten-Text-Recognizer

Handwritten text Recognizer is an OCR application that aims to recognize handwritten english sentences written in cursive. It is implemented in Pytorch through fine-tuned Resnet archictures and LSTM for sequence processing and trained using CTC loss.

Table of Contents

• Installation
• Usage
• Visualisation
• Features
• Authors
• License

Installation

Clone the repo:

git clone https://github.com/LingFengJ/Handwritten-Text-Recognizer

Install Python dependencies

 pip install -r Handwritten-Text-Recognizer/requirements.txt

Usage

Dataset

After installation, download the sentence dataset from IAM Handwriting Database
Or
From the following Kaggle Link

unzip in the data folder with the name iam_sentences

How to run the project:

• You can download our Pretrained Model
• Train your own model:

cd path/to/Handwritten-Text-Recognizer
python train.py

Visualisation

Testing:

python testing_.py
tensorboard --logdir=path/to/Handwritten-Text-Recognizer/Results

The Character Error Rate(CER) and Word Error Rate(WER) will be visible on Tensorboard

Features

A short presentation about the project

Authors

We are a group of bachelor students of Applied Computer Science and Artificial Intelligence at the Sapienza University of Rome. The Handwritten-Text-Recognizer is a project that belongs to our academic curriculum - it is designed for fullfill exam requirements of the examination "AI Lab: Computer Vision and NLP".
The process is stimulating and we benefited a lot by tackling challenges in sequential learning and deep neural network architecture design. For any clarification or further information regarding the project, please fell free to reach out to us.

• Lingfeng Jin
• Abduazizkhon Shomansurov
• Liyu Jin
• Gioia Zheng

License

Handwritten-Text-Recogniser is released under MIT License